Coreference information guides human expectations during natural reading

Proceedings of the 28th International Conference on Computational Linguistics, pages 4587–4599 Barcelona, Spain (Online), December 8-13, 2020 4587 Coreference information guides human expectations during natural reading Evan Jaffe Cory Shain William Schuler The Ohio State University {jaffe.59,shain.3,schuler.77}@osu.edu Abstract Models of human sentence processing effort tend to focus on costs associated with retrieving structures and discourse referents from memory (memory-based) and/or on costs associated with anticipating upcoming words and structures based on contextual cues (expectation-based) (Levy, 2008). Although evidence suggests that expectation and memory may play separable roles in language comprehension (Levy et al., 2013), theories of coreference processing have largely focused on memory: how comprehenders identify likely referents of linguistic expressions. In this study, we hypothesize that coreference tracking also informs human expectations about upcoming words, and we test this hypothesis by evaluating the degree to which incremental surprisal measures generated by a novel coreference-aware semantic parser explain human response times in a naturalistic self-paced reading experiment. Results indicate (1) that coreference information indeed guides human expectations and (2) that coreference effects on memory retrieval may exist independently of coreference effects on expectations. Together, these findings suggest that the language processing system exploits coreference information both to retrieve referents from memory and to anticipate upcoming material. 1 Introduction Maintaining a model of discourse referents and their relations is a core function of human language understanding and depends on the ability to recognize when linguistic expressions corefer. Identifying the referent of a linguistic expression (coreference resolution) is a well-established task in natural language processing, and a sizeable psycholinguistic literature has explored the computations that support coreference resolution in humans. Theories of coreference processing generally agree on a critical role played by searching associative memory stores for plausible referents (Greene et al., 1992; Grosz et al., 1995; Gordon, 1998; Almor, 1999; Ariel, 2001), based on findings that coreference is more easily established to more salient or activated referents, as evidenced by human reading latencies (Gordon, 1998; Cunnings et al., 2014). Nonetheless, some coreference-related phenomena are not obviously related to search. Consider the following example: Sally frightened her brother because Sally/she/he... Numerous studies indicate that he in the boldfaced slot is more difficult to process (i.e. read more slowly) than she despite the fact that he corefers unambiguously to her brother, the most recently activated discourse referent (Garvey and Caramazza, 1974; Garnham et al., 1996; Stewart et al., 2000; Koornneef and Van Berkum, 2006; Hartshorne, 2014, inter alia). This is due to the implicit causality of the verb frightened, which predisposes comprehenders to expect an explanation in terms of the agent of the frightening event, rather than the patient (Garvey and Caramazza, 1974). In addition, the full name Sally is more difficult to process than the pronoun she, despite the fact that both expressions unambiguously identify their referent. This is due to a phenomenon known as the repeated name penalty (Gordon et al., 1993). All of these expressions unambiguously identify their target referents, and for some, the preferred This work is licensed under …

Content truncated to 500 words